Wearable devices such as smartwatches are widely used as sensor nodes for IoT applications. With hardware advancement providing enough performance, these devices can host a portion of the execution of IoT applications. However, due to the strict energy budget constraints, the quality of service (QoS) of IoT applications is limited to wearable devices. In order to address this issue, recent studies have leveraged the error resiliency of IoT applications to relax the accuracy to gain performance and energy efficiency. Although gains with approximation are trivial, each IoT application has a specific limitation for accuracy, which should be smartly used to gain maximum performance and energy efficiency. Applying approximation at different phases of sensing, computation, and transmission in a pipelined manner for IoT applications exposes different choices for fine-grained accuracy-energy trade-off space. In this work, we explore the possibilities for approximating IoT applications running on resource-constrained wearable devices by characterizing energy profiles exclusively at each of the sense, compute and transmit phases. Our analysis showed variable significance in terms of energy gains upon approximating applications at different phases, which span over a range of up to 25% gains at sense phase, 88% in the compute phase and 67% at the transmit phase.